Deephos

Deephos performs predicted spectral database searches to identify and quantify TMT-labeled phosphopeptides by predicting MS/MS fragment ion spectra with deep learning and improving false discovery rate estimation via decoy spectrum generation.

Key Features:

• Predicted Spectral Database (pSDB): Compiles a pSDB from approximately 8,000 human phosphoproteins annotated in UniProt, enriched with TMT-labeled phosphopeptides for predicted fragmentation patterns.
• Deep learning-based fragment ion prediction: Uses deep learning models to predict complex fragment ion spectra resulting from TMT labeling and phosphorylation, enhancing identification sensitivity and specificity of MS/MS spectra.
• False Discovery Rate (FDR) estimation: Implements an alternative decoy spectrum generation method to provide more reliable FDR estimates in pSDB searches.
• Integrated search strategy: Combines pSDB searches with conventional database searches to enable comprehensive identification and quantification of phosphopeptides.

Scientific Applications:

• Phosphoproteome analysis in cancer: Applied to multi-stage analyses of phosphoproteomes from glioblastoma, acute myeloid leukemia, and breast cancer.
• Post-translational modification studies: Improves detection and characterization of phosphorylation on TMT-labeled peptides in mass spectrometry datasets.
• Large-scale proteogenomics: Facilitates comprehensive quantification and identification of phosphopeptides in large-scale proteogenomic studies.

Methodology:

Predicts fragment ion spectra for TMT-labeled phosphopeptides using deep learning; compiles a predicted spectral database from ~8,000 UniProt-annotated human phosphoproteins; generates decoy spectra for FDR estimation; performs predicted spectral database searches of MS/MS spectra.